Hello everyone,
This morning I had an idea: there are a lot of WISPs here from all over the world, and each one have a different PtMP configuration, so… Why don’t we share our configurations?
I think it could be an helpful thread because everyone can take the best from each configuration!
I start with mine!
I have a central user manager with rate limits on it as a RADIUS server.
Users can authenticate using PPPoE, and on the AP is automatically created the simple queue to limit them.
I have a router on each tower.
On the router I have attached APs (I use NetMetal or BaseBox).
Each AP is configured:
Mode: ap bridge
Band: 5GHz-only-N
Channel Width: 20MHz
Protocol: nv2
Bridge Mode: enabled
Multicast Helper: default
Queues
default-small type is pfifo with queue size: 30 packets.
No parent queue. All queues priority 8.
I cannot have more than 25-30 users, and then performance go very low low low low
I would like to have more users, I offer 10Mbps to each user in download e 1Mbps in upload.
Fwiw we found long ago that pfifo gives poor results shaping customer traffic, so we use sfq (one queue per customer, all children of a queue that models the backhaul capacity).
Me too, I opened a thread (http://forum.mikrotik.com/t/nv2-pfifo-low-speed-sfq-high-speed-why/75713/1) about that, but with 6.33.5 and wireless-cm2, sfq performs bad, and pfifo with 30-50 packets is better.
How much users do you have on a sector?
Do you think it can help to add a queue that models the backhaul capacity (or the sector aggregated capacity)?
Interesting. We don’t use wireless-cm2. I’m surprised the wireless interface makes any difference to this because simple queuing should not be attached to any interface queue.
You need some parent queue that constrains the total throughput over all the users because otherwise it will be possible for some number of users together to overrun the backhaul. When that happens you no longer have control over the shaping – packets are not queuing in the per-user queues and so can not be shaped according to your policy. It is not possible to fully model a sector (or really the backhaul) because the available queue policies can’t account for things like half-duplex transmission and TDD (would be nice if this were available – it is probably possible, just doesn’t exist).
